Northern Prairie Wildlife Research Center
Distribution of Fishes in the Red River of the North Basin on Multivariate
Environmental Gradients
Stream Similarity
Cluster analysis
The presence of 76 naturally reproducing stream fish species was recorded in
binary (presence or absence) format for each of 27 streams in the Red River
basin. To determine similarity of streams, the 76x27 array was used to calculate
Jaccard's coefficients (JI) (Ludwig and Reynolds 1988) for each stream pair
(A and B), where JI=a/(a+b+c), with a as the number of species that streams
A and B have in common, b as the number of species present in stream A but absent
from stream B, and c as the number of species present in stream B but absent
from stream A. The JI ranged from near 0 (for a stream pair highly dissimilar
with respect to fish species) to near 1 (stream pair very similar).
To place streams into meaningful groups, the JI for all combinations of
stream pairs were summarized into a 27x27 array or similarity matrix. An agglomerative
clustering technique (weighted centroid) provided in the Multivariate Statistical
Package (MVSP) of Kovach (1993) was used to produce a dendrogram containing
all 27 streams. A minimum JI of 0.0 was used for defining clusters (A-D).
Other measures of similarity and also measures of distance between streams
were attempted along with several different methods of clustering (single-linkage
and complete-linkage techniques). All techniques provided similar results,
with the JI and weighted centroid clustering being most meaningful ecologically.
Significance of clusters
To determine if clusters of streams produced in the CA were significantly different
from one another, a PCA (Ludwig and Reynolds 1988) was conducted followed by
a multiple response permutation procedure (MRPP) (Biondini et al. 1988). The
PCA was run using a centered covariance matrix created from the 76x27 binary
array using MVSP (Kovach 1993). Component loadings of PCA axes were species,
and PCA scores were obtained for each of the 27 streams. An ordination diagram
of streams plotted on axes 1 and 2 was produced and compared with results of
the CA.
The MRPP is a "distribution-free" permutation technique for the analysis
of ecological data which does not require parametric assumptions of normality
(Biondini et al. 1988). Input into the MRPP program were 27 observations (streams),
with five dimensions (the PCA scores for each stream from the first five axes),
in four groups (clusters A-D from CA) of sizes eight, eight, seven, and four.
Significance of clusters was determined by examining the P-value produced
by MRPP.
Previous Section -- Mapping of Fish Distribution
Return to Contents
Next Section -- Species Association Analyses